1 00:00:00,040 --> 00:00:02,560 AI has the potential to be a great equalizer. 2 00:00:02,560 --> 00:00:04,480 It can level the playing field for people with different 3 00:00:04,480 --> 00:00:06,680 abilities. On episode 71, Tool Use, we're 4 00:00:06,680 --> 00:00:09,280 joined by Sam Julien. Sam leads developer relations 5 00:00:09,280 --> 00:00:11,960 and experience at Ryder. Today we're talking about how to 6 00:00:11,960 --> 00:00:13,600 use AI to help with executive function. 7 00:00:13,800 --> 00:00:16,040 Whether it's helping with the time tags or structuring the 8 00:00:16,040 --> 00:00:19,000 unstructured thoughts, Sam has great tips for improving the 9 00:00:19,000 --> 00:00:21,360 quality of life with AI. So please enjoy this 10 00:00:21,360 --> 00:00:24,840 conversation with Sam Julien. We're both friends with Sean 11 00:00:24,840 --> 00:00:28,520 Wang Swix. I go way back with him to when 12 00:00:28,520 --> 00:00:31,800 he was doing front end and front end Devrel stuff and that kind 13 00:00:31,800 --> 00:00:33,960 of thing. And he's always my like Canary 14 00:00:33,960 --> 00:00:37,840 in the coal mine of like what what is cool and what is up and 15 00:00:37,840 --> 00:00:40,240 coming and that kind of thing. And he has a lot of depths. 16 00:00:40,240 --> 00:00:44,320 And so I had been following, you know, I've been following his 17 00:00:44,320 --> 00:00:46,400 work for, I don't know, 10 years at this point. 18 00:00:46,400 --> 00:00:50,600 But when he started late in space, that was when I started 19 00:00:50,640 --> 00:00:53,520 sort of started taking it seriously because like I had, I 20 00:00:53,920 --> 00:00:58,880 actually ChatGPT came out very close to when my son was born is 21 00:00:58,880 --> 00:01:02,560 sort of in that same era. And so I was like in the fog of 22 00:01:02,800 --> 00:01:06,160 being a newborn dad. And so like I actually couldn't 23 00:01:06,320 --> 00:01:08,280 really do anything with AI for a while. 24 00:01:10,040 --> 00:01:12,000 And so I'd I'd been like hearing the hype about it. 25 00:01:12,160 --> 00:01:14,000 It had been on my on my radar and stuff. 26 00:01:14,000 --> 00:01:18,200 But when, when Sean kind of like started late in space and I was 27 00:01:18,200 --> 00:01:19,960 really digging into it, that's when I was like, OK, this is 28 00:01:19,960 --> 00:01:25,720 like, for sure a thing. And, and so I like reading and 29 00:01:25,720 --> 00:01:28,480 learning about all that night. So I sort of started prepping 30 00:01:28,480 --> 00:01:32,120 for what eventually became my job probably 6 or 8 months 31 00:01:32,120 --> 00:01:36,720 beforehand, like really digging into everything. 32 00:01:36,720 --> 00:01:42,120 And so so I can, I didn't come at it from a sort of like aha 33 00:01:42,120 --> 00:01:45,760 moment of actually using it. I was more of like sort of a 34 00:01:46,000 --> 00:01:51,240 anthropological intellectual, like like this seems like it's 35 00:01:51,240 --> 00:01:54,240 going to be a big thing in technology and I should learn 36 00:01:54,240 --> 00:01:56,440 about it and get on top of it and that kind of thing. 37 00:01:57,240 --> 00:02:00,440 That the aha moments came later when I was like actually using 38 00:02:00,440 --> 00:02:03,440 it and getting involved with it and getting involved with AI 39 00:02:03,440 --> 00:02:04,720 engineering and that kind of thing. 40 00:02:05,040 --> 00:02:06,400 Excellent. Yeah, Skates where the puck is 41 00:02:06,400 --> 00:02:07,960 going type. Of yeah, exactly. 42 00:02:07,960 --> 00:02:10,320 That is really what it was. And so actually what they 43 00:02:10,320 --> 00:02:13,960 encourage like new dad, new job, the AI world is just like 44 00:02:13,960 --> 00:02:16,400 non-stop going. How do you leverage AI to help 45 00:02:16,400 --> 00:02:20,000 you stay organized or kind of take that massive To Do List and 46 00:02:20,000 --> 00:02:21,600 structure in a way that it's more approachable? 47 00:02:22,520 --> 00:02:25,280 It's made a huge difference there. 48 00:02:25,320 --> 00:02:27,840 There are a few very specific things I think that it really 49 00:02:27,840 --> 00:02:32,240 helps with One is that like I'm a very external processor, like 50 00:02:32,240 --> 00:02:39,320 I, I, I think by either writing or speaking, I have to like get, 51 00:02:39,520 --> 00:02:42,080 get thoughts out there and in order to do that. 52 00:02:42,080 --> 00:02:46,560 And so I think the biggest aha moment for me has been with like 53 00:02:46,560 --> 00:02:52,640 all of the AI dictation and like voice to text, all of that. 54 00:02:52,640 --> 00:02:56,160 Like on my phone, I use an app. It's one of the whisper. 55 00:02:56,200 --> 00:02:59,400 I'm sorry for the developer that I'm blanking on the actual name, 56 00:02:59,400 --> 00:03:02,360 but it's one of the whisper apps on my phone. 57 00:03:02,360 --> 00:03:04,680 I have it. Can I have it tagged to my like 58 00:03:04,720 --> 00:03:07,080 action button on my phone? And so I can just like hold it 59 00:03:07,080 --> 00:03:10,840 down. And so I'll like brain dump 60 00:03:11,280 --> 00:03:15,480 something on my mind and then I also have it then send to my 61 00:03:15,480 --> 00:03:19,320 note, my notes app called Drafts, which is like a 62 00:03:19,320 --> 00:03:24,920 legendary Mac and iOS notes app. So I can basically just like 63 00:03:24,920 --> 00:03:29,600 brain dump everything in my brain and then it does a really 64 00:03:29,600 --> 00:03:32,000 good job of transcribing it leaps and bounds better than 65 00:03:32,000 --> 00:03:35,040 Siri ever will. Then I can just edit things 66 00:03:35,040 --> 00:03:36,600 around and like move things around. 67 00:03:36,600 --> 00:03:42,040 And because I find that I often come at a problem from like 68 00:03:42,200 --> 00:03:46,720 reverse or like I'll like a lot of times I sort of back my way 69 00:03:46,800 --> 00:03:48,800 out into of a solution, you know what I mean? 70 00:03:48,800 --> 00:03:51,200 Like I know that I know the solution, but then to explain it 71 00:03:51,200 --> 00:03:54,960 to somebody else or to like turn it into like a set of work 72 00:03:54,960 --> 00:03:57,480 items, I have to like talk my way through it. 73 00:03:57,520 --> 00:04:00,160 And so it's the AI dictation is really great for that because I 74 00:04:00,160 --> 00:04:03,760 can do that brain dump that's sort of like me just meandering 75 00:04:03,760 --> 00:04:06,800 through thinking. And then I can just move all the 76 00:04:06,800 --> 00:04:10,880 text into like make it logical for somebody else who's who's 77 00:04:10,880 --> 00:04:14,280 reading and consuming it. So that's, that's a really big 78 00:04:14,280 --> 00:04:17,200 one. Let's see as far as like to do's 79 00:04:17,200 --> 00:04:19,399 and, and things like that, Trying to think where I still 80 00:04:19,399 --> 00:04:21,920 rely very heavily on like reminders and that kind of 81 00:04:21,920 --> 00:04:24,040 thing, which really doesn't have much to do with AI at this 82 00:04:24,040 --> 00:04:25,760 point. But I think for me, a lot of it 83 00:04:25,760 --> 00:04:29,480 is just giving more space in my brain, like having AI help me 84 00:04:29,480 --> 00:04:32,040 with a lot of the like rote boilerplate. 85 00:04:32,240 --> 00:04:34,360 Like it's similar with coding, you know what I mean? 86 00:04:34,360 --> 00:04:37,200 Like, like one of the best things that I think happened 87 00:04:37,200 --> 00:04:41,960 this year was Gemini getting into Gmail because it can, it 88 00:04:41,960 --> 00:04:44,520 can take all those, you know, there's like emails you get from 89 00:04:44,520 --> 00:04:47,280 like your insurance company and stuff that's like you have to 90 00:04:47,280 --> 00:04:51,360 respond to them in like a form e-mail of like hi, attached, 91 00:04:51,360 --> 00:04:53,760 please find this document for account number, blah, blah, 92 00:04:53,760 --> 00:04:54,840 blah. You know, and it like takes 93 00:04:54,840 --> 00:04:56,160 time. And I would put that stuff off 94 00:04:56,160 --> 00:04:58,960 for like an absurd amount of time because I didn't want to 95 00:04:58,960 --> 00:05:00,680 sit down and do that like tedious work. 96 00:05:01,040 --> 00:05:04,720 And so now I can just like make Gemini draft it and then I can 97 00:05:04,720 --> 00:05:07,960 just correct it and send it. And so like, suddenly I'm 98 00:05:07,960 --> 00:05:11,680 responding a lot more to things because I can just like be done 99 00:05:11,680 --> 00:05:14,240 with it. So yeah, I think that's, that's 100 00:05:14,240 --> 00:05:18,080 a big part of it for me is just like carving out more space and 101 00:05:18,080 --> 00:05:22,720 like mentally uncluttering. And I still use like my analog 102 00:05:23,560 --> 00:05:25,160 notebook and things like that. But. 103 00:05:25,320 --> 00:05:26,600 But yeah. Yeah, no, that's great. 104 00:05:27,040 --> 00:05:30,120 I've been using Aqua Voice for my transcriber lately and I've 105 00:05:30,120 --> 00:05:32,320 I've used to use Super Whisper. I've tried a few of them and I 106 00:05:32,320 --> 00:05:35,200 just happen to settle on this one right now and it does a wrap 107 00:05:35,280 --> 00:05:36,800 at the end of the year. Like how would you been using 108 00:05:36,800 --> 00:05:38,360 it? And it started in July and it 109 00:05:38,360 --> 00:05:41,320 saved me over 24 hours, apparently, because I'm speaking 110 00:05:41,320 --> 00:05:44,680 at 195 words per minute. Some of the audience that hear 111 00:05:44,680 --> 00:05:46,680 everybody saying it's a little too fast, but, you know, it is 112 00:05:46,680 --> 00:05:49,120 what it is. But just the ability to not only 113 00:05:49,120 --> 00:05:52,160 save the time from, you know, the increased bandwidth of, of 114 00:05:52,160 --> 00:05:54,800 speaking rather than typing, but I just find it's more natural. 115 00:05:54,800 --> 00:05:57,120 And if I'm trying to be very precise with writing an e-mail, 116 00:05:57,120 --> 00:05:59,040 voice transcription doesn't really work that well for me. 117 00:05:59,040 --> 00:06:01,880 But when it's a brain dump or prompting LM, I just go with 118 00:06:01,880 --> 00:06:04,760 voice So commonly now that it's really hard to go back. 119 00:06:04,760 --> 00:06:06,720 And I'm, I'm glad that more people are finding value in it 120 00:06:06,720 --> 00:06:09,160 because it's just a new modality that we can interface with 121 00:06:09,160 --> 00:06:10,760 computers that weren't always there before. 122 00:06:11,040 --> 00:06:13,600 Yeah. I use super Whisper on my a 123 00:06:13,880 --> 00:06:19,240 computer and especially for Slack, it's just like especially 124 00:06:19,240 --> 00:06:21,360 be a super whisper. You can use local models. 125 00:06:21,360 --> 00:06:22,840 You don't have to send it to the cloud. 126 00:06:22,840 --> 00:06:27,040 So like you can you can have a local whisper model so you don't 127 00:06:27,040 --> 00:06:28,960 have to worry about privacy and that kind of thing. 128 00:06:28,960 --> 00:06:32,720 And so for work, Slack or I'm sending the bazillion messages a 129 00:06:32,720 --> 00:06:35,360 day, it's like it's just such a life changer. 130 00:06:35,360 --> 00:06:38,600 Like it makes such a difference being able to just like, dictate 131 00:06:38,600 --> 00:06:40,600 responses to people. And one thing you mentioned that 132 00:06:40,760 --> 00:06:43,080 I could absolutely relate to is when you had those emails you're 133 00:06:43,080 --> 00:06:45,880 just putting off for a while. It's just, it's one of those 134 00:06:45,880 --> 00:06:48,520 unpleasant activities. But having AI to help really can 135 00:06:48,520 --> 00:06:51,360 just like quicken the process. Do you have any other examples 136 00:06:51,360 --> 00:06:54,280 where AI can help bridge the gap between your intention and your 137 00:06:54,280 --> 00:06:57,160 action? Yeah, I think another another 138 00:06:57,160 --> 00:07:01,280 really good use case I found is brainstorming, especially with 139 00:07:01,360 --> 00:07:05,440 the like thinking modes of, of the different models, whether 140 00:07:05,440 --> 00:07:09,440 it's, you know, what was previously like O1 or O3 or just 141 00:07:09,440 --> 00:07:12,480 like in now they're sort of like the thinking toggles or you 142 00:07:12,480 --> 00:07:17,160 know, any of those like sort of extended inference models and 143 00:07:17,160 --> 00:07:20,640 that kind of thing. Like having them just kind of 144 00:07:20,640 --> 00:07:24,000 help me think through things like sort of be a rubber duck of 145 00:07:24,000 --> 00:07:28,200 like, I have this problem I'm working through. 146 00:07:28,200 --> 00:07:31,760 Like, how do you, I like, what should I do here? 147 00:07:31,760 --> 00:07:34,760 Like a lot of times I know, I know the solution that I, that I 148 00:07:34,760 --> 00:07:36,720 want to get to, but I don't know how to get there. 149 00:07:36,720 --> 00:07:40,280 And I'm finding the models do a pretty good job, you know, like 150 00:07:40,280 --> 00:07:42,160 they're, they're not, they're not perfect, but like they're 151 00:07:42,160 --> 00:07:44,200 enough. Like sometimes you don't need a 152 00:07:44,200 --> 00:07:46,840 perfect response. You just need a response, you 153 00:07:46,840 --> 00:07:50,080 know, like, and for someone like me who's like an external 154 00:07:50,080 --> 00:07:54,240 processor, like I can't always like bug my wife about like 155 00:07:54,240 --> 00:07:56,920 every random problem under the sun, you know, like she's got 156 00:07:56,920 --> 00:08:00,600 enough to worry about. So like having having just a 157 00:08:00,600 --> 00:08:04,800 model that I can like brain dump and ask it to help me like like 158 00:08:04,800 --> 00:08:08,680 identify the next actionable step, you know, or like what it 159 00:08:08,680 --> 00:08:11,560 what is a step? I think I also do that a lot 160 00:08:11,560 --> 00:08:15,920 just with like mundane things like with cooking or food prep 161 00:08:15,920 --> 00:08:18,800 or like things like that, like any, any of that were that work 162 00:08:18,800 --> 00:08:23,440 that is just like cognitively difficult for me, even if it's 163 00:08:23,440 --> 00:08:27,920 not like complex, I find that to be super helpful. 164 00:08:27,920 --> 00:08:31,240 Just like having having a model breakdown, like what are what 165 00:08:31,240 --> 00:08:33,640 are the next like actionable steps and that kind of thing. 166 00:08:34,559 --> 00:08:37,120 And of course, I mean, I'm gonna kind of like blindly follow, you 167 00:08:37,120 --> 00:08:39,360 know, I'm gonna, I'm gonna like Fact Check it and that kind of 168 00:08:39,360 --> 00:08:41,159 thing. But like, it's still really 169 00:08:41,159 --> 00:08:43,559 helpful to get me unstuck a lot, you know? 170 00:08:43,720 --> 00:08:45,760 Yeah, absolutely. And, and for anyone unfamiliar, 171 00:08:45,760 --> 00:08:47,760 rubber ducking is a practice where you just talk to a rubber 172 00:08:47,760 --> 00:08:49,160 duck. And just by speaking a problem 173 00:08:49,160 --> 00:08:51,640 out loud, you tend to solve it. And now we have AI that actually 174 00:08:51,760 --> 00:08:54,160 can respond to us and help us get to the solution faster. 175 00:08:54,320 --> 00:08:57,800 Do you ever do anything like a sign, a persona, whether it's 176 00:08:57,800 --> 00:08:59,360 like, be the skeptic, be the devil's advocate? 177 00:08:59,360 --> 00:09:02,160 Do you try to have the AI take certain angles as you brainstorm 178 00:09:02,160 --> 00:09:05,080 to fill in a gap that way occasionally? 179 00:09:05,680 --> 00:09:08,920 I don't do that as often as I should and I've I've got some 180 00:09:08,920 --> 00:09:10,720 friends who've had a lot of success with that. 181 00:09:10,720 --> 00:09:15,000 I have this friend named Manuel who is on the latent space 182 00:09:15,000 --> 00:09:18,280 discord and stuff and he's like AI call him like 1000 X 183 00:09:18,280 --> 00:09:20,880 programmer we. Actually had him on earlier and 184 00:09:20,880 --> 00:09:23,800 yeah it's crazy. Yeah, he he, he does things with 185 00:09:24,000 --> 00:09:27,160 AI coding that I just like did not think any human could come 186 00:09:27,160 --> 00:09:29,760 up with. And one of the, he gave this 187 00:09:29,760 --> 00:09:33,520 like workshop at one of the AI engineer conferences a while 188 00:09:33,520 --> 00:09:35,840 back where he talked about some of his techniques. 189 00:09:35,880 --> 00:09:41,400 And like he brought up that, you know, AI, like humans in general 190 00:09:41,400 --> 00:09:46,240 tend to solve more problems adversarially than just being 191 00:09:46,240 --> 00:09:49,640 congenial with each other. And he does all these things to, 192 00:09:49,720 --> 00:09:52,880 to make the model like most of the models are sort of trained 193 00:09:52,880 --> 00:09:55,520 to be super like amiable and friendly. 194 00:09:55,520 --> 00:09:59,240 And so he does these things like he'll, he'll tell the model that 195 00:09:59,240 --> 00:10:05,120 like his code is the cast of survivor and, and like make the 196 00:10:05,120 --> 00:10:07,880 make the like take all the functions and like put them 197 00:10:07,880 --> 00:10:10,480 against each other. And like, you know, or he'll 198 00:10:10,480 --> 00:10:14,560 like say that like everything is in a play and there's like 199 00:10:14,560 --> 00:10:17,440 introduce conflict in the play. Like he does things like that 200 00:10:17,440 --> 00:10:22,560 where it's like you're, you're using like human culture to like 201 00:10:22,560 --> 00:10:26,600 make the AI like introduce conflict and tension to the AI 202 00:10:26,600 --> 00:10:30,360 to make it perform better, which is just like a like big Galaxy 203 00:10:30,360 --> 00:10:34,160 brain move. I usually get the results I 204 00:10:34,160 --> 00:10:36,120 need. Like maybe I'm just doing like 205 00:10:36,120 --> 00:10:38,760 dumb enough work that I don't really need that. 206 00:10:38,760 --> 00:10:41,920 But like I usually am fine just sort of with the default because 207 00:10:41,920 --> 00:10:45,320 like I'll, I'll do most of the cognitive lifting with my own 208 00:10:45,320 --> 00:10:47,640 brain. I just need I just need like a 209 00:10:48,040 --> 00:10:52,080 something to bounce ideas off of and like like a wall that I can 210 00:10:52,080 --> 00:10:55,240 throw things at. The only thing I really have 211 00:10:55,240 --> 00:10:57,800 done a lot with sort of more the persona thing is with like 212 00:10:57,800 --> 00:11:00,800 editing for like. So I'm working on a a book for 213 00:11:00,800 --> 00:11:05,280 O'Reilly on on graph based rag and I don't use AIA lot for like 214 00:11:05,280 --> 00:11:09,440 the actual writing because I still think it's too AII wanted 215 00:11:09,440 --> 00:11:13,800 to be my own words, but it is really helpful for brainstorming 216 00:11:13,800 --> 00:11:14,960 and editing and that kind of thing. 217 00:11:14,960 --> 00:11:19,280 So I find that like it does a better job if I tell it like, 218 00:11:19,280 --> 00:11:23,440 OK, you're a line editor for a publishing company and you need 219 00:11:23,440 --> 00:11:25,960 to follow like these like guidelines and these things 220 00:11:25,960 --> 00:11:30,120 like, and I can sort of turn that agent into an editor. 221 00:11:30,120 --> 00:11:33,240 And I'll do the same thing with like, you know, like kind of a 222 00:11:33,240 --> 00:11:37,200 system prompt of like you're, you're like a creative 223 00:11:37,200 --> 00:11:40,520 brainstorming type person. And like, like this is what? 224 00:11:40,520 --> 00:11:43,440 But so if I need it to do something really well for a 225 00:11:43,440 --> 00:11:45,840 specific task, then I'll kind of assign that persona. 226 00:11:46,480 --> 00:11:50,120 But I think I only do that 25% of the time compared to just 227 00:11:50,480 --> 00:11:52,160 sort of the day-to-day stuff, you know? 228 00:11:52,240 --> 00:11:54,440 Yeah, when I first started getting into AI, it was almost 229 00:11:54,440 --> 00:11:56,200 like best practice you have to do every time. 230 00:11:56,200 --> 00:11:58,480 And I rarely do it these days just because the malls are so 231 00:11:58,480 --> 00:12:01,680 powerful in their default state. Yeah, it's amazing how even in 232 00:12:01,680 --> 00:12:06,600 just the last year, how like I remember even a year ago, just 233 00:12:06,600 --> 00:12:10,640 the level of prompting you would have to do and just like rounds 234 00:12:10,640 --> 00:12:12,880 of tweaking your prompts and that kind of thing. 235 00:12:12,880 --> 00:12:15,240 And it's really like this last year. 236 00:12:15,240 --> 00:12:18,480 I just feel like it has really be taken off to where I can one 237 00:12:18,480 --> 00:12:22,880 shot most things and if I go the extra mile and give some 238 00:12:22,880 --> 00:12:26,040 examples I can really like one shot most things. 239 00:12:26,080 --> 00:12:28,040 And it just improves the accessibility and like, 240 00:12:28,040 --> 00:12:30,280 inclusivity of this technology because you don't have to be a 241 00:12:30,280 --> 00:12:32,360 master of what goes in to get the ideal results. 242 00:12:32,360 --> 00:12:34,840 You can kind of just wing it. You can just verbal diarrhea 243 00:12:34,840 --> 00:12:36,560 into it, and then you tend to get good results out. 244 00:12:36,680 --> 00:12:37,640 Yeah. Exactly. 245 00:12:37,640 --> 00:12:40,840 You can just, you just have like a kind of ever present 246 00:12:41,520 --> 00:12:45,240 brainstorming partner or editor or whatever in your pocket or on 247 00:12:45,240 --> 00:12:49,000 your laptop and it just works, works really well for that. 248 00:12:49,120 --> 00:12:51,480 One thing you mentioned was how you sometimes they had to help 249 00:12:51,480 --> 00:12:52,840 you tell you what the next step is. 250 00:12:53,320 --> 00:12:55,960 What how do you kind of conflate that versus having an AI 251 00:12:55,960 --> 00:12:58,920 breakdown the entire plan? Does that kind of become too 252 00:12:58,920 --> 00:13:00,840 rigid for you? Do you find just the next step 253 00:13:00,840 --> 00:13:02,840 is kind of enough to get you unstuck or I still? 254 00:13:02,840 --> 00:13:06,360 Find and I don't know. I haven't just determined yet 255 00:13:06,360 --> 00:13:09,320 whether this is because of the way my brain works or because of 256 00:13:09,320 --> 00:13:14,120 the way models are right now. But I still find for most of the 257 00:13:14,120 --> 00:13:17,920 time the AI doesn't, it doesn't judge things of the same 258 00:13:17,920 --> 00:13:20,960 importance that I do or like find insight into the same 259 00:13:20,960 --> 00:13:22,880 things that I do. Like, I tried really hard this 260 00:13:22,880 --> 00:13:26,920 year to, to build a system around social posts. 261 00:13:26,960 --> 00:13:32,040 Like like I wanted to get to a point where I could dump some 262 00:13:32,040 --> 00:13:36,200 notes from, you know, an AI Daily Brief podcast or a latent 263 00:13:36,200 --> 00:13:40,160 space podcast or like an article or something like that and have 264 00:13:40,160 --> 00:13:45,600 it give me a good couple of social posts in my voice and 265 00:13:45,600 --> 00:13:47,840 that kind of thing. But the problem I was finding 266 00:13:47,840 --> 00:13:51,240 was that like it would use the wrong, it didn't catch the same 267 00:13:51,240 --> 00:13:54,480 insights I did when I was doing a podcast or like when I was 268 00:13:54,480 --> 00:13:56,760 listening to something or reading something, the things 269 00:13:56,760 --> 00:13:59,840 that I judge to be important or insightful, like it wouldn't 270 00:13:59,880 --> 00:14:04,000 catch like it didn't. And so it really struggled to do 271 00:14:04,000 --> 00:14:05,600 that. And so I, I was trying to like 272 00:14:05,600 --> 00:14:09,400 change the, my process to like, OK, first extract the insights 273 00:14:09,400 --> 00:14:12,440 based on like these things that I tend to find important and all 274 00:14:12,440 --> 00:14:13,640 that. And it was just like so much 275 00:14:13,640 --> 00:14:16,640 work to where it's like, it's faster for me to just do this 276 00:14:16,640 --> 00:14:20,440 myself. And so I find that's a very 277 00:14:20,440 --> 00:14:24,720 similar thing with like planning that like, it will identify, 278 00:14:24,840 --> 00:14:27,400 it's sort of like, have you ever been in a situation where like 279 00:14:27,400 --> 00:14:31,880 Google Maps, like gives you what is technically the fastest 280 00:14:31,880 --> 00:14:33,840 route, but it makes zero sense at all. 281 00:14:33,840 --> 00:14:36,960 Like, so in Portland, there's like all these rivers, like 282 00:14:36,960 --> 00:14:40,640 there's one big river that's all these bridges and nobody goes 283 00:14:40,640 --> 00:14:42,440 across the bridge unless they have to. 284 00:14:42,440 --> 00:14:45,480 Like, it's just, it's really dumb to go to like cross the 285 00:14:45,480 --> 00:14:47,480 bridges over and over again because they get so backed up 286 00:14:47,480 --> 00:14:48,840 with traffic. And when I first moved to 287 00:14:48,840 --> 00:14:51,760 Portland, Google Maps would just have me going back and forth 288 00:14:51,760 --> 00:14:54,120 across the bridges because it was technically the shortest 289 00:14:54,120 --> 00:14:56,320 distance, but it was completely illogical. 290 00:14:56,320 --> 00:14:59,200 It made no sense. And so I find that like AI does 291 00:14:59,200 --> 00:15:03,280 that a lot with planning where it's like, yes, OK, technically 292 00:15:03,280 --> 00:15:05,840 this is the next step, but like, it's really dumb to do it in 293 00:15:05,840 --> 00:15:08,560 that order because there's all these other like factors and 294 00:15:08,560 --> 00:15:10,440 contexts that you don't understand. 295 00:15:11,480 --> 00:15:15,400 And so I, I hopefully that will also change over time. 296 00:15:15,400 --> 00:15:17,280 But again, I don't know if that's like because of the way 297 00:15:17,280 --> 00:15:20,560 my brain works or because of like the way models are. 298 00:15:20,560 --> 00:15:24,800 And so I do find it to be a lot more like, it's a lot more 299 00:15:24,800 --> 00:15:29,080 effective for me to give it like to just help me get to sort of 300 00:15:29,080 --> 00:15:31,560 like what's the next. It may give me like a broad 301 00:15:31,560 --> 00:15:34,600 picture and like 60% of it will be good. 302 00:15:34,600 --> 00:15:37,160 But then if I can narrow it down and be like, OK, here's exactly 303 00:15:37,160 --> 00:15:40,480 what I'm deal with dealing with, like what's the next step in 304 00:15:40,480 --> 00:15:43,120 this process? It's it's pretty good at that. 305 00:15:43,120 --> 00:15:45,440 Yeah, I have actually countered similar issues when doing the 306 00:15:45,440 --> 00:15:47,720 descriptions for these videos where I used to just give the 307 00:15:47,720 --> 00:15:50,160 whole transcript basic prompt. It was OK. 308 00:15:50,160 --> 00:15:52,960 Tried to like prompt engineer for a while and until I found 309 00:15:52,960 --> 00:15:55,240 out just like focus on the intro because I was just kind of give 310 00:15:55,240 --> 00:15:57,280 a quick summary of what it is. And as soon as I focus on that, 311 00:15:57,280 --> 00:15:59,600 it's fine because it's just more, more distilled. 312 00:16:00,040 --> 00:16:02,360 But if you try to give it the entire thing, yeah, it just kind 313 00:16:02,360 --> 00:16:05,080 of picks random things. It gets hyper fixated on it. 314 00:16:05,080 --> 00:16:07,640 Have you noticed any other tendencies with current state of 315 00:16:07,640 --> 00:16:11,520 LMS that kind of make certain tasks just unacceptable to hand 316 00:16:11,520 --> 00:16:13,440 off to AI? Yeah, I mean, I think, I think 317 00:16:13,440 --> 00:16:16,520 it all comes down to the the whole thing of like context 318 00:16:16,520 --> 00:16:19,440 engineering right now that is like such a big prop like 319 00:16:19,880 --> 00:16:21,880 challenge. Like the models themselves, they 320 00:16:21,880 --> 00:16:24,440 think like are getting pretty good. 321 00:16:24,440 --> 00:16:28,880 But but but the, the challenge of giving the right context at 322 00:16:28,880 --> 00:16:32,320 the right moment and the right instructions at the right moment 323 00:16:32,320 --> 00:16:35,960 is very, very difficult. And so I think what I've been 324 00:16:35,960 --> 00:16:38,680 encountering, especially over this last year as the models 325 00:16:38,680 --> 00:16:42,480 have gotten so good, is like there's so much hidden context 326 00:16:42,480 --> 00:16:45,000 in your brain that you don't even realize. 327 00:16:45,360 --> 00:16:48,120 That's like because you've got your all of your sensory 328 00:16:48,120 --> 00:16:51,080 perception and all of your life experience and everything that 329 00:16:51,080 --> 00:16:57,120 like is affecting how you see a problem and try to come up with 330 00:16:57,120 --> 00:17:04,359 a solution that trying to get this blank slate or an LLM that 331 00:17:04,359 --> 00:17:08,640 has its own biases and things like that trying to arrive at a 332 00:17:08,640 --> 00:17:10,440 conclusion. So I still find it quite 333 00:17:10,440 --> 00:17:14,359 difficult for anything that's like really strategic or 334 00:17:14,359 --> 00:17:17,160 creative or things like that, which which is fine with me 335 00:17:17,160 --> 00:17:18,880 because that's the work I really want to be doing. 336 00:17:18,880 --> 00:17:22,000 Like I, I, I like we're in such a like hype state. 337 00:17:22,000 --> 00:17:25,760 And I continue to find, I said, I said this last, like in 2024 338 00:17:25,760 --> 00:17:28,000 when I was giving talks and writing articles and stuff was 339 00:17:28,000 --> 00:17:31,920 like, I feel like the biggest ROI with AI, with AI right now 340 00:17:31,920 --> 00:17:36,600 is like the most boring, like drudgery stuff like being able 341 00:17:36,600 --> 00:17:38,640 to get rid of that. Like I'm not really looking for, 342 00:17:38,640 --> 00:17:41,600 I mean, I would like AI to do things like help with curing 343 00:17:41,600 --> 00:17:45,240 cancer and things like that. But as far as like other like 344 00:17:45,960 --> 00:17:49,800 crazy amazing things, like I don't think that that's, I think 345 00:17:49,800 --> 00:17:52,720 we've actually arrived at a lot of really important ROI that 346 00:17:52,720 --> 00:17:55,840 we're just overlooking because there's just a lot of like 347 00:17:56,160 --> 00:17:58,440 boring work that AI does really well. 348 00:17:58,440 --> 00:18:00,200 And so I'm definitely finding that. 349 00:18:00,920 --> 00:18:03,360 So yeah, the biggest problems right now I think are context 350 00:18:03,960 --> 00:18:08,960 and that often effects like really complex problems or like 351 00:18:08,960 --> 00:18:12,760 creative and strategic things. But as we've seen with coding 352 00:18:12,760 --> 00:18:18,720 like that landscape is getting, I mean, it's incredible how much 353 00:18:18,720 --> 00:18:21,440 has changed there. So it's entirely possible that a 354 00:18:21,440 --> 00:18:23,960 year from now or six months from now, don't feel totally 355 00:18:23,960 --> 00:18:26,280 different about it. This flywheel effect has been 356 00:18:26,280 --> 00:18:29,200 absolutely insane. It's wonderful to see from the 357 00:18:29,200 --> 00:18:32,480 inside, but the distribution of the benefits is definitely not 358 00:18:32,640 --> 00:18:33,960 even yet. Like, there's a few people 359 00:18:33,960 --> 00:18:36,280 really benefiting and we really gotta get it out there more. 360 00:18:36,560 --> 00:18:38,640 And I think, yeah, some of the context would be helpful. 361 00:18:38,640 --> 00:18:41,960 Very big lift, but one thing you just brought up about how like 362 00:18:42,080 --> 00:18:45,000 helping reduce the the boring, the drudgery 1 area that I'm 363 00:18:45,000 --> 00:18:47,160 curious if you've thought about or explore at all, Do you think 364 00:18:47,240 --> 00:18:50,080 AI can help with burnout where it might be able to get rid of 365 00:18:50,080 --> 00:18:52,120 those things? They're just so monotonous, so 366 00:18:52,120 --> 00:18:55,280 soul crushing that we're able to be a little more fulfilled, even 367 00:18:55,280 --> 00:18:57,840 if it doesn't necessarily like replace jobs or eliminate jobs, 368 00:18:57,840 --> 00:19:00,680 it just gets rid of like a certain subset of the task that 369 00:19:00,680 --> 00:19:03,320 we don't want to do anymore. Oh yeah, 100%. 370 00:19:03,600 --> 00:19:07,880 I think one of the, one of the biggest ROI things for me, I 371 00:19:07,880 --> 00:19:12,000 mean, it's related to the, the voice stuff, but it's like AI 372 00:19:12,000 --> 00:19:13,720 note taking and that kind of thing. 373 00:19:13,720 --> 00:19:18,200 Because you know, I like any other PM or, or whoever like 374 00:19:18,200 --> 00:19:21,320 have to do weekly reports and weekly summaries and you know, 375 00:19:21,320 --> 00:19:24,560 like monthly reporting and like all that stuff, you know, like I 376 00:19:24,560 --> 00:19:26,960 have to like keep track of a bunch of stuff and like kind of 377 00:19:26,960 --> 00:19:29,200 going back to like the reminders and To Do List and all of that 378 00:19:29,200 --> 00:19:34,320 stuff. Like having a like whatever your 379 00:19:34,840 --> 00:19:38,840 AI transcription tool of choices running in a meeting that can 380 00:19:38,840 --> 00:19:42,560 capture things you committed to or things you need to remember 381 00:19:42,560 --> 00:19:45,640 or like things like that. Like that's been such a game 382 00:19:45,640 --> 00:19:49,080 changer because for me, that's such a cognitive load of like 383 00:19:49,200 --> 00:19:51,480 trying to remember that stuff and then having to like type it 384 00:19:51,480 --> 00:19:54,520 all out and, and all of that. Like instead having like a 385 00:19:55,640 --> 00:19:59,560 decent first pass at it through that, those note taking tools 386 00:19:59,560 --> 00:20:03,720 like that for me, I mean, it sounds like a small thing, but 387 00:20:03,720 --> 00:20:08,080 it's like if you're spending, I mean, at my last role, because 388 00:20:08,080 --> 00:20:10,480 my last role was at a publicly traded company. 389 00:20:10,480 --> 00:20:13,440 And so I was doing a lot of paperwork and stuff like that 390 00:20:13,440 --> 00:20:15,720 and. It all really adds up. 391 00:20:15,720 --> 00:20:17,960 And that was contributing to a lot of my burnout. 392 00:20:17,960 --> 00:20:20,960 It was just like all the admin work and all the drudgery and 393 00:20:20,960 --> 00:20:22,320 stuff. Like I wanted to be working with 394 00:20:22,320 --> 00:20:24,800 my team and doing creative work, but I was spending a lot of time 395 00:20:24,800 --> 00:20:28,800 doing paperwork, you know, and AI can help with a lot of that 396 00:20:28,800 --> 00:20:30,880 now. And I'm definitely noticing that 397 00:20:30,880 --> 00:20:34,320 like this, having those like several hours a week, like maybe 398 00:20:34,320 --> 00:20:38,600 an hour a day of that note taking and reporting and admin 399 00:20:38,600 --> 00:20:42,800 work taken off my shoulders as like a huge, huge benefit now. 400 00:20:42,920 --> 00:20:45,120 Yeah, and I found the exact same thing. 401 00:20:45,160 --> 00:20:49,160 I use granola personally, but it's not just the time save from 402 00:20:49,160 --> 00:20:51,160 not taking notes. I can be way more present in a 403 00:20:51,160 --> 00:20:53,080 call instead of just like trying to like make mental notes. 404 00:20:53,160 --> 00:20:55,360 Yeah, yeah. I just like fully talk, forget 405 00:20:55,360 --> 00:20:57,640 about things going weird tangents and it'll capture 406 00:20:57,640 --> 00:20:59,400 everything. You can query it after. 407 00:20:59,400 --> 00:21:01,840 You can hook up a pipeline to bring into your To Do List as 408 00:21:01,840 --> 00:21:03,880 you were mentioning. So it's just one of those like 409 00:21:04,200 --> 00:21:06,360 background processes that most people don't really think about, 410 00:21:06,360 --> 00:21:08,960 but it just allows the quality of life, like the quality of a 411 00:21:08,960 --> 00:21:11,680 job, to get so much better. Yeah, 100%. 412 00:21:11,760 --> 00:21:14,440 And it's, it's again, it's like, I think the theme that I found 413 00:21:14,440 --> 00:21:20,000 is just like, it's the, the small, like daily compounding 414 00:21:20,000 --> 00:21:25,360 things that that really add up and either contribute to having 415 00:21:25,360 --> 00:21:28,040 a really successful time or, or like burning out. 416 00:21:28,040 --> 00:21:31,520 You know, like, I think, I think we tend to assume that success 417 00:21:31,520 --> 00:21:34,440 or burnout is like in all the big swings you take. 418 00:21:34,440 --> 00:21:37,000 And like, to some extent it is like, obviously if you're like 419 00:21:37,000 --> 00:21:39,280 in some super toxic environment, you're going to burn out. 420 00:21:39,280 --> 00:21:42,680 Or like, if you like do some crazy award-winning thing, 421 00:21:42,680 --> 00:21:43,920 you're going to like be successful. 422 00:21:43,920 --> 00:21:47,080 But like, I think for most people, most of the time, it's 423 00:21:47,080 --> 00:21:50,880 like what you do every single day and like the 1% progress 424 00:21:50,880 --> 00:21:57,080 you're making every day or 1% attraction is really like what 425 00:21:57,080 --> 00:21:59,960 the what is going to contribute one way or another, you know? 426 00:22:00,000 --> 00:22:01,640 Yeah. And actually on that note, do 427 00:22:01,640 --> 00:22:04,640 you have any other micro skills that you would like to share 428 00:22:04,640 --> 00:22:07,040 with the audience for things that might not be like a big 429 00:22:07,040 --> 00:22:09,200 game changer by themselves, but they do compound over time to 430 00:22:09,200 --> 00:22:10,560 help make quality of life better? 431 00:22:10,600 --> 00:22:12,800 I've been sort of on this journey with like AI and 432 00:22:12,800 --> 00:22:15,520 everything and like I was, I was writing about AI for a little 433 00:22:15,520 --> 00:22:20,600 while because I think there's this big gap in AI engineering 434 00:22:20,600 --> 00:22:24,000 skills right now that a lot of developers need help with and 435 00:22:24,000 --> 00:22:26,000 that kind of thing. But some of some of them, I 436 00:22:26,000 --> 00:22:30,760 think my best work was really in the like anti hype productivity 437 00:22:31,240 --> 00:22:35,280 space, sort of like really practical things that are 438 00:22:35,280 --> 00:22:38,160 actually really helpful. And so like the whole like 1% 439 00:22:38,160 --> 00:22:41,600 compounding thing, that was like something I wrote a while ago 440 00:22:43,440 --> 00:22:46,560 and like identifying the next action and that kind of thing. 441 00:22:47,560 --> 00:22:50,760 And so I think, I think it's like looking for opportunities 442 00:22:50,760 --> 00:22:55,120 that you can do small experiments on things and, and 443 00:22:55,120 --> 00:22:56,840 like test them out. I think one of the best 444 00:22:56,840 --> 00:22:59,880 microscales I picked up on over the last decade was like this 445 00:22:59,880 --> 00:23:03,920 concept that like it's not a failure, it's a test like 446 00:23:04,520 --> 00:23:07,880 looking, seeing, seeing life as sort of like a feedback loop. 447 00:23:08,280 --> 00:23:11,240 And if you're sucking at something, it probably is an 448 00:23:11,240 --> 00:23:13,360 indication that you're like getting experience and have an 449 00:23:13,360 --> 00:23:15,680 opportunity to get better. I literally like just thought 450 00:23:15,680 --> 00:23:17,640 about this the other day. It was something about being a 451 00:23:17,640 --> 00:23:20,440 parent where I was like, I'm really struggling with this and 452 00:23:20,440 --> 00:23:23,920 like like that, that hopefully that means I'm like getting 453 00:23:23,920 --> 00:23:27,600 experience, you know? And so I think things like that, 454 00:23:27,600 --> 00:23:30,920 that like AI doesn't actually fundamentally change, you know, 455 00:23:30,920 --> 00:23:34,840 about the human experience where there's like small things that 456 00:23:34,840 --> 00:23:40,160 you can do to like improve your your life or your career or 457 00:23:40,160 --> 00:23:42,680 things like that, that don't necessarily have to be big and 458 00:23:42,680 --> 00:23:44,720 dramatic, you know? And just on the note of that, 459 00:23:44,720 --> 00:23:48,800 the productivity angle, I found a habit that fortunately I had 460 00:23:48,800 --> 00:23:50,360 before. I was just being curious where I 461 00:23:50,360 --> 00:23:53,480 can open up a chat with Claude, Say I have this one workflow. 462 00:23:53,480 --> 00:23:55,680 I would like you to generate a script that helps with it and 463 00:23:55,920 --> 00:23:57,280 just kind of explore if possible there. 464 00:23:57,280 --> 00:23:59,720 I like, I use a Mac and I have hammer spoons so I can get 465 00:23:59,920 --> 00:24:02,480 access to the API. I have a few different command 466 00:24:02,480 --> 00:24:05,520 line tools, just little things that I encountered, which I 467 00:24:05,520 --> 00:24:07,880 don't necessarily need to use AI in the actual practice of it, 468 00:24:08,040 --> 00:24:11,120 but just AI allowed me to gain the knowledge of how to leverage 469 00:24:11,120 --> 00:24:13,040 these existing tools without having to go through the whole 470 00:24:13,040 --> 00:24:15,800 learning process normally. So just having the willingness 471 00:24:15,800 --> 00:24:18,520 to just ask questions and have this conversations, be curious, 472 00:24:18,520 --> 00:24:21,760 get messy, all all those fun things I had, It really just UPS 473 00:24:21,760 --> 00:24:24,080 quality of life because you get these benefits that you didn't, 474 00:24:24,160 --> 00:24:26,000 you weren't even originally intending to get. 475 00:24:26,000 --> 00:24:29,040 You just asked some questions. Yeah, yeah, for sure. 476 00:24:29,720 --> 00:24:33,760 I've even like, I've, I've had a lot of like RSI and hand and 477 00:24:33,760 --> 00:24:35,880 wrist issues and things over over the years. 478 00:24:35,880 --> 00:24:40,000 And like now I'm, I, I use like Chad TPT and Claude to help 479 00:24:40,240 --> 00:24:43,760 troubleshoot some of that stuff of like not that you should like 480 00:24:43,760 --> 00:24:46,720 use it for like medical stuff, but like, you know, I'm having 481 00:24:46,720 --> 00:24:50,120 pain in this and I can like snap a picture of like my setup and 482 00:24:50,120 --> 00:24:53,480 be like, why is this happening? Like how can I adjust my, my 483 00:24:53,480 --> 00:24:56,120 setup? And like, it helped me find like 484 00:24:56,120 --> 00:25:00,720 a, a couple of like assistive devices to use and, and that 485 00:25:00,720 --> 00:25:03,560 kind of thing. Like, yeah, even like even like 486 00:25:03,560 --> 00:25:07,080 home repair stuff where it's like, if they're like, I think 487 00:25:07,080 --> 00:25:10,920 so much, no matter what the topic is, I think there's so 488 00:25:10,920 --> 00:25:14,520 much inertia of just figuring out what the first step is. 489 00:25:14,880 --> 00:25:19,240 You know, of like if I have a, a home project or, or, or a coding 490 00:25:19,240 --> 00:25:21,600 project or whatever where I just like, I have no idea how to 491 00:25:21,600 --> 00:25:24,280 start. Like the micro skill is like 492 00:25:24,280 --> 00:25:27,320 just start just like pick something and do it and, and get 493 00:25:27,320 --> 00:25:29,600 feedback and everything. But sometimes the, the 494 00:25:29,600 --> 00:25:32,200 overwhelming feeling of like, I don't know where to start is 495 00:25:32,200 --> 00:25:36,680 like enough to stop. And I found like AI has been 496 00:25:36,680 --> 00:25:40,280 very helpful for that of just like in writing, we have the 497 00:25:40,280 --> 00:25:41,840 phrase of like the shitty first draft. 498 00:25:41,840 --> 00:25:44,280 Like you just have to like get something out there. 499 00:25:44,280 --> 00:25:46,760 And I think there's also sort of like a parallel, like the shitty 500 00:25:46,760 --> 00:25:50,080 first step, like just like do something even even if it's 501 00:25:50,080 --> 00:25:52,120 terrible. And so like you can kind of use 502 00:25:52,120 --> 00:25:55,360 AI to just like get you there of like, I have a problem. 503 00:25:55,760 --> 00:25:58,360 What's what's the answer? And usually it's somewhat wrong 504 00:25:58,360 --> 00:26:00,160 because you didn't really describe the problem well. 505 00:26:00,160 --> 00:26:03,600 And then that is like enough of a catalyst to like get you 506 00:26:03,600 --> 00:26:05,960 moving, you know, and get and get you doing things. 507 00:26:06,040 --> 00:26:08,440 Yeah, break the inertia. Like unclog the pipe, just get 508 00:26:08,440 --> 00:26:11,080 going one area. And I don't have any personal 509 00:26:11,080 --> 00:26:13,480 experience in this, but I am curious if you've either heard 510 00:26:13,480 --> 00:26:17,200 or like come across it yourself where now with AI kind of 511 00:26:17,200 --> 00:26:19,880 unlocking all these things and giving us all of this. 512 00:26:20,160 --> 00:26:23,240 Once you break the inertia, all this exposure, is there a risk 513 00:26:23,240 --> 00:26:26,280 of getting like decision fatigue or just having too many choices? 514 00:26:26,280 --> 00:26:28,720 Is it, is it potentially it could be counterproductive and 515 00:26:28,720 --> 00:26:30,920 just unlocking too many aspects that we just kind of get 516 00:26:30,920 --> 00:26:32,840 overwhelmed going the other way? I do. 517 00:26:32,920 --> 00:26:35,920 I do think so because it's like, I feel like I've experienced 518 00:26:35,920 --> 00:26:39,320 this on a few levels. Like from a work standpoint, it, 519 00:26:39,320 --> 00:26:42,080 it is tough because now the timelines are a lot more 520 00:26:42,080 --> 00:26:44,680 compressed and the possibilities are a lot more extended. 521 00:26:44,680 --> 00:26:48,760 And so it's, it's harder, I think to manage expectations and 522 00:26:48,760 --> 00:26:52,480 set boundaries and like be really clear about your capacity 523 00:26:52,480 --> 00:26:57,240 and things like that. Because now you know, some an 524 00:26:57,240 --> 00:27:00,560 update to the docs that would take a week in prior years now 525 00:27:00,560 --> 00:27:03,400 takes like 30 seconds with play with, you know what I mean? 526 00:27:03,480 --> 00:27:07,720 Like with, with cursor or like we use Mintlify and they've, 527 00:27:07,760 --> 00:27:11,200 they've got this new agent that's just mind blowingly good 528 00:27:12,640 --> 00:27:14,840 and like, you know, writing an article or things like that. 529 00:27:14,840 --> 00:27:17,920 And so it, it does make it a little difficult because it's 530 00:27:17,920 --> 00:27:21,000 like, OK, I could, I could, I could spend all day just like 531 00:27:21,000 --> 00:27:24,760 flooding the Internet with like Dachshund content and things 532 00:27:24,760 --> 00:27:27,040 like that. But like, should I like what? 533 00:27:27,040 --> 00:27:28,640 Like is that a good use of my time? 534 00:27:28,640 --> 00:27:31,040 Like I think that's a really good thing to have to think 535 00:27:31,040 --> 00:27:34,720 through. So I do think there is like that 536 00:27:34,720 --> 00:27:39,800 danger there of like that. I also think in, in tandem with 537 00:27:39,800 --> 00:27:42,360 that is like an issue that everybody is sort of talking 538 00:27:42,360 --> 00:27:45,000 about is like the whole echo chamber phenomenon of like, I 539 00:27:45,000 --> 00:27:51,240 think it's also possible to, if you're only talking with these 540 00:27:51,280 --> 00:27:54,760 AI products that are basically like incentivized to make you 541 00:27:54,760 --> 00:27:57,720 feel good. Like it's very easy to get into 542 00:27:57,720 --> 00:28:02,120 an echo Chamber of like drinking your own kool-aid and, and just 543 00:28:02,120 --> 00:28:05,440 like thinking that like you're solving all these problems and, 544 00:28:05,440 --> 00:28:08,320 and kind of go off in a ditch. And like, it's really important 545 00:28:08,320 --> 00:28:11,240 that you have like actual, whether it's like customer 546 00:28:11,240 --> 00:28:15,120 feedback or friends or, you know, medical professionals or 547 00:28:15,120 --> 00:28:18,920 lawyers or whatever, like who can like, you know, give you 548 00:28:19,160 --> 00:28:23,120 actual, real feedback. You know, that that like you can 549 00:28:23,280 --> 00:28:26,240 break out of that. You know, like, I think about 550 00:28:26,240 --> 00:28:29,120 that a lot with my, my son, because it's like he's going to 551 00:28:29,120 --> 00:28:31,320 grow up in this world. There's actually this huge 552 00:28:31,720 --> 00:28:35,960 article in the Economist recently about how AI is 553 00:28:35,960 --> 00:28:37,880 affecting childhood and, and things like that. 554 00:28:37,880 --> 00:28:42,880 And, and that's a big fear is like how if, if kids start using 555 00:28:42,960 --> 00:28:48,080 AI for studying and chatting and all of that, like, and they're 556 00:28:48,080 --> 00:28:50,760 not going to be forced to develop all of these like social 557 00:28:50,760 --> 00:28:53,480 skills and like cognitive reasoning and like things like 558 00:28:53,480 --> 00:28:56,240 that. So I do think it is there's that 559 00:28:56,240 --> 00:28:58,520 whole, you know, people's joke about like the smooth brain 560 00:28:58,520 --> 00:29:01,760 syndrome and stuff like that of like just brain rot and that 561 00:29:01,760 --> 00:29:04,240 kind of thing. So I think there's it's a double 562 00:29:04,240 --> 00:29:05,880 edged sword. You know, there's like all the 563 00:29:05,880 --> 00:29:08,200 possibilities and also all the, all the, all the ways that you 564 00:29:08,200 --> 00:29:12,000 can sort of like, 'cause your own demise if you're not careful 565 00:29:12,000 --> 00:29:13,880 with it, you know? Yeah, no, absolutely. 566 00:29:14,120 --> 00:29:17,320 Just last weekend actually, I was on a panel in Waterloo for 567 00:29:17,320 --> 00:29:19,680 youth tech labs and I kind of put together a thing about AI. 568 00:29:19,680 --> 00:29:21,840 Where do you go from here? And there were six of us on a 569 00:29:21,840 --> 00:29:24,360 panel. And the opinions of AI reach 570 00:29:24,360 --> 00:29:26,640 from miraculous all the way to like extreme fear. 571 00:29:26,800 --> 00:29:28,600 So it's really cool looking different perspectives, which I 572 00:29:28,600 --> 00:29:30,800 think is important. But afterwards we had breakout 573 00:29:30,800 --> 00:29:33,280 sessions with some of the different students and we were 574 00:29:33,280 --> 00:29:35,600 tasked with just kind of like, what are their thoughts? 575 00:29:35,600 --> 00:29:38,800 Like how how is AI affecting their life as like a 14 to 18 576 00:29:38,800 --> 00:29:40,520 year old? And it's really interesting 577 00:29:40,520 --> 00:29:42,040 here. Some of them talk about how 578 00:29:42,200 --> 00:29:44,360 because of all the the AI generated slot, they're actually 579 00:29:44,360 --> 00:29:47,600 less inclined to want to keep scrolling and go online, which 580 00:29:47,600 --> 00:29:49,040 is something I didn't anticipate at all. 581 00:29:49,400 --> 00:29:52,680 And they were even worried that when you're super interested in 582 00:29:52,680 --> 00:29:55,000 the subject, you can just like go down a rabbit hole for 583 00:29:55,000 --> 00:29:56,800 Infinity. You can learn it so deeply. 584 00:29:56,800 --> 00:30:00,240 So they're very bright, but that the resilience of having to go 585 00:30:00,240 --> 00:30:03,000 through like writing an English essay on a book you find boring. 586 00:30:03,160 --> 00:30:05,640 And just like grinder project is something that that's what those 587 00:30:05,640 --> 00:30:07,760 go to ChatGPT. So it's this weird thing where 588 00:30:07,760 --> 00:30:11,360 they're getting like more capable terms of intellect, but 589 00:30:11,360 --> 00:30:13,960 almost like less capable to terms of resilience because it's 590 00:30:13,960 --> 00:30:16,960 just this enabling tool. Yeah, I think it's going to be 591 00:30:16,960 --> 00:30:22,000 very important that we have other outlets to teach that and 592 00:30:22,000 --> 00:30:24,320 other other ways to do that. You know, where you're in the, 593 00:30:24,640 --> 00:30:28,080 in the physical world or you're in, in the like the, in the 594 00:30:28,080 --> 00:30:31,320 Economist, they're talking about shifting more emphasis on to in 595 00:30:31,320 --> 00:30:35,560 classroom writing and in classroom like debating and 596 00:30:35,560 --> 00:30:36,800 lecturing and that kind of thing. 597 00:30:36,800 --> 00:30:40,040 So that there is no strategy. BT you can just dump that into. 598 00:30:40,040 --> 00:30:43,040 Or, you know, I think about like real world, you know, physical 599 00:30:43,040 --> 00:30:46,400 skills. Like I, I take when I was in a 600 00:30:46,400 --> 00:30:49,080 classic Portland move, I took a bunch of like blacksmithing and 601 00:30:49,080 --> 00:30:51,960 welding classes when I moved to Portland And like that took, 602 00:30:51,960 --> 00:30:54,640 taught me a lot of life lessons about, you know, literal 603 00:30:54,640 --> 00:30:58,160 grinding, you know, and like having to like suffer 3 things. 604 00:30:58,160 --> 00:31:01,600 And you know, if you for those of us who were learning coding 605 00:31:01,680 --> 00:31:05,080 before AI, like a lot of your best experience comes from like 606 00:31:05,080 --> 00:31:07,520 the worst problems you had to do, you know, like the most 607 00:31:07,520 --> 00:31:11,320 boring, annoying problems you had to do in coding, you 608 00:31:11,320 --> 00:31:14,200 actually learn the most, you know, because you're like just 609 00:31:14,320 --> 00:31:17,000 banging your head against it and trying to like understand it, 610 00:31:17,000 --> 00:31:20,440 you know, at a deeper level. So yeah, it is, it is really 611 00:31:20,440 --> 00:31:22,800 interesting. But yeah, I do think about, you 612 00:31:22,800 --> 00:31:26,280 know, even now, like being able to learn anything, like even 613 00:31:26,280 --> 00:31:30,120 like last night, I like was scrolling TikTok and there was 614 00:31:30,120 --> 00:31:34,560 this really cool, like Japanese cover of a song. 615 00:31:34,920 --> 00:31:38,320 And I was able to just like, screenshot it and get the figure 616 00:31:38,320 --> 00:31:40,920 out what song it was. And it let me down this whole 617 00:31:40,920 --> 00:31:43,760 rabbit hole of this, like, Japanese emo band that I had 618 00:31:43,760 --> 00:31:46,400 never heard of. And was like, like digging into 619 00:31:46,400 --> 00:31:48,800 the, like, lyrics and like getting the translations and 620 00:31:48,800 --> 00:31:50,520 like, and it was like incredible, you know, I just 621 00:31:50,520 --> 00:31:53,040 like went down this whole rabbit hole of like, I would never been 622 00:31:53,040 --> 00:31:57,160 able to do that, you know, like, so like that kind of thing is, 623 00:31:57,160 --> 00:31:59,560 is super fun now, you know, And I think it's like, yeah, if I'd 624 00:31:59,560 --> 00:32:02,880 had that in school and the ability to like learn anything 625 00:32:02,880 --> 00:32:05,960 at any time, very, you know, like go deep into whatever, 626 00:32:05,960 --> 00:32:07,840 like, yeah, that's incredible, you know. 627 00:32:08,000 --> 00:32:10,200 It's funny you mentioned the the grind process because I still 628 00:32:10,200 --> 00:32:13,120 remember early days of coding. 2 full work days, not solving a 629 00:32:13,120 --> 00:32:15,600 problem, showing up to work, leaving frustrated twice in a 630 00:32:15,600 --> 00:32:16,720 row. Incredible. 631 00:32:16,720 --> 00:32:18,760 Like if Stack Overflow didn't have the answer, you just have 632 00:32:18,760 --> 00:32:21,520 to figure it out. But once too, one thing I do 633 00:32:21,520 --> 00:32:25,120 want to touch on is when we're doing a lot of writing and AI 634 00:32:25,120 --> 00:32:27,640 helpers like frame it in this way or make it appropriate for 635 00:32:27,640 --> 00:32:29,000 this, make it more professional or what not. 636 00:32:29,360 --> 00:32:32,520 And I kind of worry that because so much our experience now 637 00:32:32,520 --> 00:32:35,760 online and especially like AI filtered, we're not necessarily 638 00:32:35,760 --> 00:32:39,800 developing the skills to like modify your own behavior for 639 00:32:39,800 --> 00:32:42,520 social interactions. Do you see AI helping with that 640 00:32:42,520 --> 00:32:45,400 or hindering that? I think right now we're going in 641 00:32:45,400 --> 00:32:47,560 a, a direction that it's hindering and I think we have to 642 00:32:47,560 --> 00:32:52,320 be very proactive about putting things in place to prevent that 643 00:32:52,320 --> 00:32:55,000 because I do think we're like we, we already, because of the 644 00:32:55,000 --> 00:32:57,680 way technology has globalized things, we're sort of going in 645 00:32:57,680 --> 00:33:03,480 this like monoculture direction. I like, and, and I think I do 646 00:33:03,480 --> 00:33:05,480 see that a lot with like writing, You know, if you look 647 00:33:05,480 --> 00:33:08,360 at like, it's so easy to identify AI writing at this 648 00:33:08,360 --> 00:33:11,280 point and like everything sort of sounds the same. 649 00:33:11,280 --> 00:33:14,520 And it's because it's because it's just basically this like 650 00:33:15,720 --> 00:33:18,600 mush of all the writing that's on the Internet, you know, and 651 00:33:18,600 --> 00:33:21,120 it's sort of like the lowest common denominator. 652 00:33:21,120 --> 00:33:25,760 And so like, I do think we really need to be careful to not 653 00:33:26,240 --> 00:33:30,000 let that translate into like culture and music and, you know, 654 00:33:30,000 --> 00:33:32,600 social interaction and personality types and all of 655 00:33:32,600 --> 00:33:34,480 that stuff. Like we can't sort of like let 656 00:33:35,000 --> 00:33:42,160 let the like training data just like start affecting us in that 657 00:33:42,160 --> 00:33:44,640 way. You know, I, I have some hope 658 00:33:44,640 --> 00:33:48,360 that that I think I think there is going to be this swing and I 659 00:33:48,360 --> 00:33:50,920 think we're sort of already seeing it this sort of like 660 00:33:51,880 --> 00:33:56,080 reaction to a lot of the the swing into AI where there's 661 00:33:56,080 --> 00:34:01,120 going to be more of a desire for art and poetry and music and 662 00:34:01,600 --> 00:34:06,920 language and like analog, you know, records and notebooks and 663 00:34:06,920 --> 00:34:08,480 things like that. Like I think there's going to be 664 00:34:08,480 --> 00:34:11,719 a strong movement for that, you know, and I think that's really 665 00:34:11,719 --> 00:34:12,719 good. I think people are going to 666 00:34:12,719 --> 00:34:15,480 really want to double down on like what makes us human and 667 00:34:16,400 --> 00:34:18,440 human connection and that kind of thing. 668 00:34:18,440 --> 00:34:22,120 You know, like, I think it's, I think there's something innate 669 00:34:22,360 --> 00:34:26,280 like in our evolution that, that wants that, you know, I can see 670 00:34:26,280 --> 00:34:30,080 it with my own child that like, I mean, he doesn't do anything 671 00:34:30,080 --> 00:34:32,920 with AI at this point. But like, just with, even with 672 00:34:32,920 --> 00:34:38,440 technology, like he could have all the toys around him he could 673 00:34:38,440 --> 00:34:42,520 ever wish for and get bored. But I stick him outside and he 674 00:34:42,520 --> 00:34:45,040 can run around and like play in leaves and in the grass and 675 00:34:45,040 --> 00:34:46,360 stuff. He could be out there for hours 676 00:34:46,360 --> 00:34:49,960 just like moving rocks from one part of the yard to the other, 677 00:34:49,960 --> 00:34:53,440 you know, or, you know, and singing and, and all of that, 678 00:34:53,440 --> 00:34:55,679 you know, like it's, I think there's just something in us 679 00:34:55,679 --> 00:34:58,800 that like, I think it's the same, I think that's the same 680 00:34:58,800 --> 00:35:02,280 phenomenon of when you open Sora and you have no desire to like 681 00:35:02,640 --> 00:35:05,200 scroll endlessly in it. That the, that, you know, the, 682 00:35:05,560 --> 00:35:08,120 the folks you were talking to is like, I had that same feeling as 683 00:35:08,120 --> 00:35:10,680 like, this is weird. And I don't, you know, like, I 684 00:35:10,760 --> 00:35:12,280 like, I have no desire to do this. 685 00:35:12,480 --> 00:35:15,880 Whereas like on TikTok, I'm seeing like other people's 686 00:35:15,880 --> 00:35:18,720 perspectives and, you know, like a window into their life and 687 00:35:18,720 --> 00:35:19,880 stuff. And that's like super 688 00:35:19,880 --> 00:35:23,160 fascinating to me, you know, like, so yeah, hopefully, 689 00:35:23,160 --> 00:35:26,160 hopefully that's sort of the half the glass half full of 690 00:35:26,160 --> 00:35:29,360 Roach, I guess of like humans will overcome that. 691 00:35:30,120 --> 00:35:32,920 Yeah, I'm, I'm optimistic. I'm hopeful as well. 692 00:35:34,120 --> 00:35:37,520 I fully agree that like in real life, IRL events are going to 693 00:35:37,520 --> 00:35:40,440 become more valuable as people to crave that interaction. 1 694 00:35:40,440 --> 00:35:43,360 funny point on your point of the monoculture. 695 00:35:43,360 --> 00:35:46,000 I still remember when I was teaching English in Korea, one 696 00:35:46,000 --> 00:35:48,760 of the students at Lol and I was like, what, what does that mean? 697 00:35:48,760 --> 00:35:50,840 Like teacher, it's funny. I'm like, no, no, but what does 698 00:35:50,840 --> 00:35:51,920 it stand for? Like I have no idea. 699 00:35:51,920 --> 00:35:54,440 And Internet culture just permeates all other cultures. 700 00:35:54,680 --> 00:35:57,040 And I think diversity of culture is extremely important. 701 00:35:57,320 --> 00:36:00,800 And I do worry that if AIS are sole interface to digital world 702 00:36:00,800 --> 00:36:02,600 like that, that variety goes away. 703 00:36:02,920 --> 00:36:06,120 So I really hoping that AI will enable us to do, you know, put 704 00:36:06,120 --> 00:36:08,680 down the screen and step back, go touch grass and actually like 705 00:36:08,680 --> 00:36:11,200 interact with people while, you know, the background processes 706 00:36:11,200 --> 00:36:13,120 are handling our bills and and the boring stuff. 707 00:36:13,560 --> 00:36:16,360 Yeah, that's the hope. Sam, this was a blast. 708 00:36:16,360 --> 00:36:18,280 I really enjoyed talking to you. Before we let you go, is there 709 00:36:18,280 --> 00:36:19,440 anything you'd like the audience to know? 710 00:36:19,640 --> 00:36:22,320 No, I don't think so. I do. 711 00:36:22,360 --> 00:36:26,720 I still do my writing on cmglean.com and I'm always 712 00:36:27,360 --> 00:36:30,960 fascinated to have people reach out and talk to me about any of 713 00:36:30,960 --> 00:36:34,920 these, any of these topics. But yeah, I know it's it's a 714 00:36:34,920 --> 00:36:37,920 wild ride. And yeah, we really enjoyed the 715 00:36:37,920 --> 00:36:40,760 conversation. And yeah, thanks for having me. 716 00:36:41,160 --> 00:36:42,520 Absolutely. And actually, before I got to 717 00:36:42,520 --> 00:36:45,160 make sure a lot of our audience would care about your graph rag 718 00:36:45,160 --> 00:36:46,480 book, do you have any idea when? That's cool. 719 00:36:46,600 --> 00:36:49,320 Yeah. My wife is always like like we 720 00:36:49,320 --> 00:36:51,240 were, we were just at a family gathering, you know, for 721 00:36:51,240 --> 00:36:54,240 Thanksgiving and everything and like my brother-in-law was 722 00:36:54,240 --> 00:36:55,480 asking me like what I'm up to and stuff. 723 00:36:55,480 --> 00:36:58,280 And he was like, and you're writing a book And I was like, 724 00:36:58,320 --> 00:36:59,920 Oh yeah, yeah, I'm writing a book. 725 00:37:01,680 --> 00:37:02,200 Yeah. No. 726 00:37:02,200 --> 00:37:04,040 Which is actually like a dream come true. 727 00:37:04,200 --> 00:37:08,920 I'm I'm I'm Co authoring a book with my friend Anthony Alcaraz, 728 00:37:08,920 --> 00:37:13,880 who's AI person. I don't even know what his title 729 00:37:13,880 --> 00:37:20,760 is, but he's just an AI guru. I don't know in Paris, but we're 730 00:37:20,760 --> 00:37:26,120 writing a book called agentic graph rag and it's basically a 731 00:37:26,120 --> 00:37:31,320 book for O'Reilly on how graphs can influence all the different 732 00:37:31,320 --> 00:37:36,280 parts of agent architecture. It's, it's a really interesting 733 00:37:36,280 --> 00:37:41,200 topic and so we we're Co authoring it and it, it should 734 00:37:41,200 --> 00:37:43,880 be, we're finishing the manuscript sometime in the first 735 00:37:43,880 --> 00:37:47,120 quarter of 2026. And so it should be published 736 00:37:47,120 --> 00:37:49,240 sometime in the summer of of 2026. 737 00:37:49,240 --> 00:37:52,040 And when we were putting early release chapters on. 738 00:37:52,040 --> 00:37:55,360 And so I think there's a chapter on there that goes over the the 739 00:37:55,400 --> 00:38:00,400 agent architecture and then some early, an early chapter on 740 00:38:00,400 --> 00:38:02,080 agentic memory and that kind of thing. 741 00:38:02,080 --> 00:38:06,760 So yeah, graph based rag is sort of like there's it's in some 742 00:38:06,760 --> 00:38:09,880 ways it's it's like a hot topic and in other ways it's not. 743 00:38:09,880 --> 00:38:14,840 But like, for example, most of the AI memory tools now like ZAP 744 00:38:14,840 --> 00:38:17,520 and Cogni and Mem Zero and all those things, those those are 745 00:38:17,520 --> 00:38:20,200 all graph based. And so there's some really, 746 00:38:20,200 --> 00:38:23,480 really interesting stuff happening in with graph based 747 00:38:23,480 --> 00:38:24,960 rag and agents and that kind of thing. 748 00:38:24,960 --> 00:38:27,440 And so just a really interesting topic to write about. 749 00:38:27,440 --> 00:38:30,320 So definitely check it out. It's it's at oreilly.com if you 750 00:38:30,400 --> 00:38:32,240 if you look for it. Yeah, I'll make sure to link it 751 00:38:32,240 --> 00:38:34,200 down below and and hopefully get to have you back on when it's 752 00:38:34,200 --> 00:38:35,600 out so we can talk more in depth about that. 753 00:38:35,920 --> 00:38:38,520 Yeah, definitely love to love to talk more about that. 754 00:38:38,640 --> 00:38:40,760 Thank you for this. My conversation with Sam, Julie. 755 00:38:41,160 --> 00:38:43,400 Sam and I were joking after he finished recording the episode 756 00:38:43,400 --> 00:38:45,520 how nice it was to not talk about work and be able to just 757 00:38:45,520 --> 00:38:48,480 explore the the fun parts of AI, the things that actually improve 758 00:38:48,480 --> 00:38:51,160 quality of life. You don't necessarily have to 759 00:38:51,160 --> 00:38:54,320 take every single tool that's out there or try to be hyper 760 00:38:54,320 --> 00:38:57,720 productive all the time. You can find the aspects of your 761 00:38:57,720 --> 00:39:01,080 life that add friction, add drudgery, are just boring and 762 00:39:01,080 --> 00:39:03,880 see if AI can help with that. If you just have a little chat 763 00:39:03,880 --> 00:39:06,600 with Claude or ChatGPT or whatever your model of choices 764 00:39:06,600 --> 00:39:09,760 and just say, this is the part of my day that I hate the most, 765 00:39:09,960 --> 00:39:12,280 how can I make it better? You'll probably find something. 766 00:39:12,520 --> 00:39:14,800 And if you work on these micro skills, as Sam calls them, 767 00:39:15,000 --> 00:39:17,720 you're able to compound over time and actually get to a spot 768 00:39:17,720 --> 00:39:19,680 that you probably didn't think was possible to begin with. 769 00:39:20,040 --> 00:39:22,520 You don't have to necessarily adhere to the structure of 770 00:39:22,520 --> 00:39:25,560 modern society where you can just be yourself and have AI be 771 00:39:25,560 --> 00:39:27,960 the translation layer. I hope you enjoy this episode 772 00:39:27,960 --> 00:39:29,080 and I'll see you next week.